Student choices, prompt, assignment, audience, and required concepts.
Make the important parts of AI creation inspectable.
The Glass Box is designed to replace “AI magic” with a student-friendly view of what can actually be observed: inputs, sources, tools, system stages, generated claims, and verification evidence.
Explore one generation from five angles.
Use the tabs to inspect a hypothetical photosynthesis story.
What did the student tell AI?
Create an 8-page Grade 5 science adventure about Maya shrinking down and exploring the inside of a leaf.
Transparency is useful only when it supports a decision.
Teacher materials, approved references, evidence snippets, and provenance.
Major observable steps such as policy application, retrieval, generation, and review.
Claims labeled by support status instead of being presented as equally reliable.
Student tasks that require opening evidence and making a judgment.
Operational transparency.
- Actual student inputs and assignment parameters
- Actual retrieved source references where source grounding is used
- Actual system event categories recorded by the platform
- Post-generation claim analysis and support labels
- Student verification actions and reflections
A claim to read an AI’s mind.
- Not hidden chain-of-thought
- Not “the AI’s secret thoughts”
- Not a guarantee that every supported claim is correct
- Not a substitute for student or teacher judgment
- Not a confidence meter disguised as certainty
Build the explanation from system metadata, not from a model inventing a story afterward.
The production design should record events such as request received, classroom policy applied, source retrieved, content generated, claims extracted, claims checked, safety review completed, and output delivered. The Glass Box can then summarize real system activity in age-appropriate language.